Summary of Auto-train-once: Controller Network Guided Automatic Network Pruning From Scratch, by Xidong Wu et al.
Auto-Train-Once: Controller Network Guided Automatic Network Pruning from Scratch
by Xidong Wu, Shangqian Gao, Zeyu Zhang, Zhenzhen Li, Runxue Bao, Yanfu Zhang, Xiaoqian Wang, Heng Huang
First submitted to arxiv on: 21 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Auto-Train-Once (ATO) algorithm is a novel network pruning method that eliminates the need for additional fine-tuning steps by directly training and compressing general deep neural networks (DNNs) from scratch. ATO leverages a controller network as an architecture generator to guide the learning of target model weights during training, while also utilizing a novel stochastic gradient algorithm to enhance coordination between model training and controller network training. This approach is shown to achieve state-of-the-art performance across various model architectures (ResNet18, ResNet34, ResNet50, ResNet56, and MobileNetv2) on standard benchmark datasets (CIFAR-10, CIFAR-100, and ImageNet). |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to make deep neural networks smaller and more efficient. Normally, you need to train the network multiple times to get it to work well, but this approach trains the network just once and then makes it smaller. The method uses a special controller network that helps the main network learn in the right way. This leads to better results than other methods for making networks smaller. The paper shows that this approach works well on many different types of networks and datasets. |
Keywords
* Artificial intelligence * Fine tuning * Pruning
